Ookla Net Index (based on Speedtest, Pingtest results)

Forums › New Zealand Broadband › Ookla Net Index (based on Speedtest, Pingtest results)

freitasm

BDFL - Memuneh
80658 posts

Uber Geek
+1 received by user: 41071

Administrator

ID Verified

Trusted

Geekzone

Lifetime subscriber

#69390 7-Oct-2010 09:50

Now all those speedtests you have been performing can be put into nice charts - covering costs, speeds, etc.

Here is the Wellington entrypoint (click in My Location to change the page).

TelstraClear performance could be so much higher if they peered...

Referral links: Quic Broadband (free setup code: R587125ERQ6VE) | Samsung | AliExpress | Wise | Sharesies

Support Geekzone by subscribing (browse ads-free), or making a one-off or recurring donation through PressPatron.

muppet

2644 posts

Uber Geek
+1 received by user: 1662

Trusted

#389070 7-Oct-2010 10:13

Why will peering make a difference though?

Sure - they don't peer. But almost any ISP with clue buys a domestic circuit off either TelstraClear or Telecom. The fact they had to purchase this circuit (i.e. pay money) is exactly why TelstraClear/TCom don't peer. They make money because they can force people to buy domestic circuits off them.

The thing is, it doesn't actually affect network performance if everyone does buy a circuit into one or the other. The network paths are there, you're just having to pay for them.

If you buy a Domestic circuit from Telecom, you also get access to TelstraClear's domestic traffic too.

So the only reason "not peering" could potentially be a _performance_ problem is if the domestic circuit that an ISP buys isn't bit enough and gets congested. Which is probably the case in a few instances. Or of course if and ISP doesn't buy a domestic circuit at all and then traffic will route "Internationally" This doesn't necessarily mean it leaves the country, it might, but it might not depending on how the routing works.

The network topology itself isn't a problem. It's just that people are having to pay to get access to TelstraClear and Telecom, instead of peering "for free".
The problem is a financial one, not really a technical one.

In summary, peering vs buying a circuit is a commerical problem. But packets don't care if they link they're flowing over costs money or not.

Tim

Audiophiles are such twits! They buy such pointless stuff: Gold plated cables, $2000 power cords. Idiots.

OOOHHHH HYPERFIBRE!

freitasm

BDFL - Memuneh
80658 posts

Uber Geek
+1 received by user: 41071

Administrator

ID Verified

Trusted

Geekzone

Lifetime subscriber

#389073 7-Oct-2010 10:17

In real life peering is not a problem, because as you say any self respecting ISP would do anything to peer with Telecom and TelstraClear, the two largest ISPs in the country.

For the effects of these tests peering is a problem though because most users in Wellington automatically click to test using the Wellington server, which is hosted by Citylink, which is not a peering partner.

So in reality those results are skewed. This is my point.

I am using Truenet here for testing. When the test is done in their primary server the results show 6Mbps... When done to their secondary service it shows almost 30Mbps.

See where I am getting at?

Those testings, while a good reference are not authoritative.

Referral links: Quic Broadband (free setup code: R587125ERQ6VE) | Samsung | AliExpress | Wise | Sharesies

Support Geekzone by subscribing (browse ads-free), or making a one-off or recurring donation through PressPatron.

freitasm

BDFL - Memuneh
80658 posts

Uber Geek
+1 received by user: 41071

Administrator

ID Verified

Trusted

Geekzone

Lifetime subscriber

#389076 7-Oct-2010 10:22

muppet: Why will peering make a difference though?

Because the routing will be different - in the case of TelstraClear it means actually going through a route that is not the most direct one. Sometimes they even have packets going overseas before coming back to a site next to the client.

For example:

See the difference? Logic would tell us the Wellington server would be faster, because it's just around the corner from me. But it's not what happens. And now those test results are stored in Ookla's database...

Referral links: Quic Broadband (free setup code: R587125ERQ6VE) | Samsung | AliExpress | Wise | Sharesies

Support Geekzone by subscribing (browse ads-free), or making a one-off or recurring donation through PressPatron.

muppet

2644 posts

Uber Geek
+1 received by user: 1662

Trusted

#389094 7-Oct-2010 11:05

freitasm:
muppet: Why will peering make a difference though?

Because the routing will be different - in the case of TelstraClear it means actually going through a route that is not the most direct one.

Traffic always flows the most direct route (unless you're doing policy based routing) - it's just in this case the most direct route is a crap one. Performance would also be improved if CityLink were buying a domestic connection, or if the size of their link in the US was increased.

But CityLink don't buy a domestic circuit and thus all traffic from TelstraClear/Telecom to them goes via their rather small International gateway in the US.

So yes, I agree that performance for this specific speedtest would be improved if TelstraClear (or Telecom) peered with CityLink. I thought that you were saying overall Internet performance on TelstraClear would be improved if they peered, which isn't true.

On a side note I think it's great that CityLink don't buy a circuit off Telecom/TelstraClear. CityLink has content that TelstraClear/TCom customers want and their own stupidity (TCom's/TClear's, not citylink) by not peering means their customers get crap performance for anything CityLink hosted.
I wish everyone would drop their domestic circuits, leaving TCom and TelstraClear as their own little Internet routing island. But that'd mean using end-users as pawns so that's never going to happen.

freitasm: Logic would tell us the Wellington server would be faster, because it's just around the corner from me.

When it comes to routing, this is flawed logic.

Audiophiles are such twits! They buy such pointless stuff: Gold plated cables, $2000 power cords. Idiots.

OOOHHHH HYPERFIBRE!

muppet

2644 posts

Uber Geek
+1 received by user: 1662

Trusted

#389103 7-Oct-2010 11:20

That wasn't prehaps the best "I was wrong, you were right" I've ever posted, and I hate it when people won't admit they were wrong. So, just to be clear:

If TelstraClear peered, the Speedtest results you've posted would be much higher.

Audiophiles are such twits! They buy such pointless stuff: Gold plated cables, $2000 power cords. Idiots.

OOOHHHH HYPERFIBRE!

freitasm

BDFL - Memuneh
80658 posts

Uber Geek
+1 received by user: 41071

Administrator

ID Verified

Trusted

Geekzone

Lifetime subscriber

#389105 7-Oct-2010 11:21

muppet: That wasn't prehaps the best "I was wrong, you were right" I've ever posted, and I hate it when people won't admit they were wrong. So, just to be clear:

If TelstraClear peered, the Speedtest results you've posted would be much higher.

Correct. But it doesn't fix the broken Internet "steps" we have here in New Zealand, which is somethig we both agree with, from the looks of it.

Referral links: Quic Broadband (free setup code: R587125ERQ6VE) | Samsung | AliExpress | Wise | Sharesies

Support Geekzone by subscribing (browse ads-free), or making a one-off or recurring donation through PressPatron.

Geekzone

Want to support Geekzone and browse the site without the ads? Subscribe to Geekzone now (monthly, annual and lifetime options).

Ragnor

8279 posts

Uber Geek
+1 received by user: 585

Trusted

#389186 7-Oct-2010 14:12

Ookla speedtest and pingtest is pretty useless now for gauging real world performance.

An ISP' would be foolish not to white list traffic to their servers, I bet most do now.

freitasm

BDFL - Memuneh
80658 posts

Uber Geek
+1 received by user: 41071

Administrator

ID Verified

Trusted

Geekzone

Lifetime subscriber

#389187 7-Oct-2010 14:12

I heard exactly the same thing yesterday during a meeting...

Referral links: Quic Broadband (free setup code: R587125ERQ6VE) | Samsung | AliExpress | Wise | Sharesies

Support Geekzone by subscribing (browse ads-free), or making a one-off or recurring donation through PressPatron.

Ragnor

8279 posts

Uber Geek
+1 received by user: 585

Trusted

#389191 7-Oct-2010 14:17

Project B Grade is a little more interesting as it does 2x national tests and 2x international but isn't very active anymore
http://bgrade.bage.net.nz/

You can schedule their client exe to run every hour if you want an to see how your connection does over time.

Same problem eventually though - unless they keep changing the test download urls, ISP's will whitelist them, however if you use real world addresses like steam client download etc that's actually a positive
result!

http://bgrade.bage.net.nz/client/client_sources.php?format=html

muppet

2644 posts

Uber Geek
+1 received by user: 1662

Trusted

#389197 7-Oct-2010 14:26

Ragnor: Ookla speedtest and pingtest is pretty useless now for gauging real world performance.

An ISP' would be foolish not to?white list?traffic to their servers, I bet most do now.

I think you could drop the work "pretty" and just flat out say it is useless. It measures speed between two points only. In New Zealand, with the exception of the posted TClear->CityLink graph, it almost always travels over a peering connection. So you end up with speeds like this:

All that's done is run over my peering link though. I don't have that sort of bandwidth over my Intl' link.

Speedtest will show you if your line/backhaul is congested, but it won't give any real performance stats. Which is why I tend to quietly snicker at people that have them in their sigs...

Project B Grade sounds interesting, I'll have to have a look at it. I prefer to just run smokeping to ~30 different destinations. You can get a decent idea if your ISPs links are becoming congested because the ping times tend to creep up during peak.

I also love what SamKnows has done.

Audiophiles are such twits! They buy such pointless stuff: Gold plated cables, $2000 power cords. Idiots.

OOOHHHH HYPERFIBRE!

freitasm

BDFL - Memuneh
80658 posts

Uber Geek
+1 received by user: 41071

Administrator

ID Verified

Trusted

Geekzone

Lifetime subscriber

#389224 7-Oct-2010 15:23

How is SamKnows different from TrueNet (of which I have a box and posted some early results here)?

You can see from my results the differences are still there...

Referral links: Quic Broadband (free setup code: R587125ERQ6VE) | Samsung | AliExpress | Wise | Sharesies

Support Geekzone by subscribing (browse ads-free), or making a one-off or recurring donation through PressPatron.

AliExpress

Shop now on AliExpress (affiliate link).

muppet

2644 posts

Uber Geek
+1 received by user: 1662

Trusted

#389229 7-Oct-2010 15:35

freitasm: How is SamKnows different from?TrueNet (of which I have a box and posted some early results here)?

You can see from my results the differences are still there...

Doesn't seem to be much different. I assumed TrueNet was an ISP in your first post. I can't see any per-ISP stas on their page, but no doubt they have them.

Audiophiles are such twits! They buy such pointless stuff: Gold plated cables, $2000 power cords. Idiots.

OOOHHHH HYPERFIBRE!

freitasm

BDFL - Memuneh
80658 posts

Uber Geek
+1 received by user: 41071

Administrator

ID Verified

Trusted

Geekzone

Lifetime subscriber

#389235 7-Oct-2010 15:38

They have. I only posted a screenshot of my specific connection. I was shown stats for multiple ISPs yesterday when I met then here in Wellington.

You get a DLink box with a custom firmware to run the tests. In my case it's not my primary router as I am using a Small Business Cisco router with my VoIP lines on it. But the DLink is plugged directly into the router and tests should be ok.

As I said in my other post, TrueNet is looking for more volunteers.

Referral links: Quic Broadband (free setup code: R587125ERQ6VE) | Samsung | AliExpress | Wise | Sharesies

Support Geekzone by subscribing (browse ads-free), or making a one-off or recurring donation through PressPatron.

Ragnor

8279 posts

Uber Geek
+1 received by user: 585

Trusted

#389346 7-Oct-2010 21:07

Yeah but the problem is if you want to get a realistic picture you need run 5-10 different tests (that might consume 50MB) say every hour that works aoubt about 1GB of traffic per day for "testing" or 30GB per month.

On our capped plans, who's going to want to do that.

Alternatively if they are only testing will small amounts of data/files they are going to get fairly biased results as a lot of management only kicks in on streams/connections larger than certain sizes.

With that said it's nice to see some Kiwi's doing it, maybe they will be able to get win the contract from the ComCom off Epitiro.